KINSMCSuppose you want an illustration of a blue mug beside a pear. You have a photo with an arrangement you like and a painting with the right colours and texture. They solve different problems. In Leonardo AI, use Content Reference when the arrangement matters and Style Reference when you want to borrow the visual treatment. A short comparison with the same prompt will show you which mode gets you closer for the image you have in mind.
Choose the reference by what you want to preserve
Style Reference helps guide the way an image looks. Think about a restrained colour palette, a grainy print finish or the shading of a soft pencil drawing. To test it, make a reference yourself with a few swatches and a patch of texture. If you use a photo of a different scene, decide which visual traits you want and spell them out in the prompt. The generated scene may pick up other details from the reference, so look at the result with fresh eyes.
Content Reference helps guide the shapes and structure of an image. Try a simple still-life photo you took yourself, with a mug on the left and a pear on the right. The reference gives the model a visual starting point for those forms. For a site header, it can be helpful to include the empty area you want to reserve for a title in the reference itself. Keep that requirement in the prompt too. Check the resulting crop before placing any text.
Leonardo also documents a separate Character Reference option for likeness. Use the manufacturer’s compatibility guidance to see which reference controls your selected model offers. If a control is unavailable, check the model and settings before assuming you missed a menu. For your first comparison, pick a model that supports the two options you want to try.
Run a comparison you can judge in five minutes
- Write one prompt. For instance, ask for an editorial illustration of a blue ceramic mug and a green pear on a cream tabletop, with empty space above them for a heading.
- Set a baseline. Use the same model, dimensions and generation settings for each run. Generate an image without a reference so you can see how the prompt alone is interpreted.
- Change one input. Run the prompt with your swatches as a Style Reference. Then use your still-life photo as a Content Reference with the same prompt. Start with the default reference strength if the interface gives you a choice, and write down what you used.
- Judge each result against the brief. Record whether it has the right palette, object arrangement and space for the heading. Keep a thumbnail of each result next to your notes.
Those first runs test two practical ways to use your materials, but both the source image and reference mode differ. For a closer comparison, use the same reference image with each mode and hold the other settings steady. You might use a still life that has both a clear layout and a distinctive palette. Repeat a run when one result looks like an outlier, since generation can vary between attempts.
Troubleshoot one visible problem at a time
If the style is right but the pear keeps moving behind the mug, describe the positions more clearly or try a simpler layout reference. If the layout is close but the surface looks too glossy, use a style image that shows the matte finish plainly. When the interface offers reference strength, try a small adjustment and save the result with its settings. Changing the prompt, model and strength together makes it harder to tell which change helped.
Before using the finished image, inspect it at the size it will appear. Look for unexpected lettering, logos, distorted edges and the space needed for a title. Use a reference you created or have permission to use, and avoid uploading photos that contain private details. If you plan to make repeated images for the same site, keep your brief, references, model choice and preferred settings with the approved output so you can reproduce the process later.
Start with the constraint you would reject an image for getting wrong. For a set of article covers, that might be the colour treatment. For a banner that needs room for a title, it might be composition. If you are editing a finished image after generation, our AI image editing workflow explains how to describe the change and preserve the parts that already work.